4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 based upon some outcommented c code from mpeg2dec (idct_mmx.c
22 written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
26 #include "simple_idct.h"
29 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */
30 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */
31 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */
32 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */
33 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */
34 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */
35 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */
39 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
40 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
41 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
42 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
43 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
44 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
45 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
47 #define COL_SHIFT 20 // 6
50 /* 8x8 Matrix used to do a trivial (slow) 8 point IDCT */
51 static int coeff[64]={
52 W4, W4, W4, W4, W4, W4, W4, W4,
53 W1, W3, W5, W7,-W7,-W5,-W3,-W1,
54 W2, W6,-W6,-W2,-W2,-W6, W6, W2,
55 W3,-W7,-W1,-W5, W5, W1, W7,-W3,
56 W4,-W4,-W4, W4, W4,-W4,-W4, W4,
57 W5,-W1, W7, W3,-W3,-W7, W1,-W5,
58 W6,-W2, W2,-W6,-W6, W2,-W2, W6,
59 W7,-W5, W3,-W1, W1,-W3, W5,-W7
62 static int inline idctRowCondZ (int16_t * row)
64 int a0, a1, a2, a3, b0, b1, b2, b3;
66 if( !( ((uint32_t*)row)[0]|((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3])) {
67 /* row[0] = row[1] = row[2] = row[3] = row[4] =
68 row[5] = row[6] = row[7] = 0;*/
72 if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){
73 a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1));
74 a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1));
75 a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1));
76 a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1));
78 b0 = W1*row[1] + W3*row[3];
79 b1 = W3*row[1] - W7*row[3];
80 b2 = W5*row[1] - W1*row[3];
81 b3 = W7*row[1] - W5*row[3];
83 a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1));
84 a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1));
85 a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1));
86 a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1));
88 b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7];
89 b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7];
90 b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7];
91 b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7];
94 row[0] = (a0 + b0) >> ROW_SHIFT;
95 row[1] = (a1 + b1) >> ROW_SHIFT;
96 row[2] = (a2 + b2) >> ROW_SHIFT;
97 row[3] = (a3 + b3) >> ROW_SHIFT;
98 row[4] = (a3 - b3) >> ROW_SHIFT;
99 row[5] = (a2 - b2) >> ROW_SHIFT;
100 row[6] = (a1 - b1) >> ROW_SHIFT;
101 row[7] = (a0 - b0) >> ROW_SHIFT;
107 /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */
108 static int inline idctRowCondDC(int16_t *row)
110 int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;
111 uint64_t *lrow = (uint64_t *) row;
116 if ((lrow[0] & ~0xffffULL) == 0) {
120 a0 += 1 << (ROW_SHIFT - 1);
158 a0 += 1 << (ROW_SHIFT - 1);
159 a1 += 1 << (ROW_SHIFT - 1);
160 a2 += 1 << (ROW_SHIFT - 1);
161 a3 += 1 << (ROW_SHIFT - 1);
196 row[0] = (a0 + b0) >> ROW_SHIFT;
197 row[1] = (a1 + b1) >> ROW_SHIFT;
198 row[2] = (a2 + b2) >> ROW_SHIFT;
199 row[3] = (a3 + b3) >> ROW_SHIFT;
200 row[4] = (a3 - b3) >> ROW_SHIFT;
201 row[5] = (a2 - b2) >> ROW_SHIFT;
202 row[6] = (a1 - b1) >> ROW_SHIFT;
203 row[7] = (a0 - b0) >> ROW_SHIFT;
207 #else /* not ARCH_ALPHA */
208 static int inline idctRowCondDC (int16_t * row)
210 int a0, a1, a2, a3, b0, b1, b2, b3;
212 if( !( ((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3]| row[1])) {
213 // row[0] = row[1] = row[2] = row[3] = row[4] = row[5] = row[6] = row[7] = row[0]<<3;
214 uint16_t temp= row[0]<<3;
215 ((uint32_t*)row)[0]=((uint32_t*)row)[1]=
216 ((uint32_t*)row)[2]=((uint32_t*)row)[3]= temp + (temp<<16);
220 if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){
221 a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1));
222 a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1));
223 a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1));
224 a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1));
226 b0 = W1*row[1] + W3*row[3];
227 b1 = W3*row[1] - W7*row[3];
228 b2 = W5*row[1] - W1*row[3];
229 b3 = W7*row[1] - W5*row[3];
231 a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1));
232 a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1));
233 a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1));
234 a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1));
236 b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7];
237 b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7];
238 b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7];
239 b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7];
242 row[0] = (a0 + b0) >> ROW_SHIFT;
243 row[7] = (a0 - b0) >> ROW_SHIFT;
244 row[1] = (a1 + b1) >> ROW_SHIFT;
245 row[6] = (a1 - b1) >> ROW_SHIFT;
246 row[2] = (a2 + b2) >> ROW_SHIFT;
247 row[5] = (a2 - b2) >> ROW_SHIFT;
248 row[3] = (a3 + b3) >> ROW_SHIFT;
249 row[4] = (a3 - b3) >> ROW_SHIFT;
253 #endif /* not ARCH_ALPHA */
255 static void inline idctCol (int16_t * col)
259 if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
260 col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
261 col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
265 int a0, a1, a2, a3, b0, b1, b2, b3;
266 col[0] += (1<<(COL_SHIFT-1))/W4;
267 a0 = W4*col[8*0] + W2*col[8*2] + W4*col[8*4] + W6*col[8*6];
268 a1 = W4*col[8*0] + W6*col[8*2] - W4*col[8*4] - W2*col[8*6];
269 a2 = W4*col[8*0] - W6*col[8*2] - W4*col[8*4] + W2*col[8*6];
270 a3 = W4*col[8*0] - W2*col[8*2] + W4*col[8*4] - W6*col[8*6];
272 b0 = W1*col[8*1] + W3*col[8*3] + W5*col[8*5] + W7*col[8*7];
273 b1 = W3*col[8*1] - W7*col[8*3] - W1*col[8*5] - W5*col[8*7];
274 b2 = W5*col[8*1] - W1*col[8*3] + W7*col[8*5] + W3*col[8*7];
275 b3 = W7*col[8*1] - W5*col[8*3] + W3*col[8*5] - W1*col[8*7];
277 col[8*0] = (a0 + b0) >> COL_SHIFT;
278 col[8*7] = (a0 - b0) >> COL_SHIFT;
279 col[8*1] = (a1 + b1) >> COL_SHIFT;
280 col[8*6] = (a1 - b1) >> COL_SHIFT;
281 col[8*2] = (a2 + b2) >> COL_SHIFT;
282 col[8*5] = (a2 - b2) >> COL_SHIFT;
283 col[8*3] = (a3 + b3) >> COL_SHIFT;
284 col[8*4] = (a3 - b3) >> COL_SHIFT;
287 static void inline idctSparseCol (int16_t * col)
289 int a0, a1, a2, a3, b0, b1, b2, b3;
290 col[0] += (1<<(COL_SHIFT-1))/W4;
352 col[8*0] = (a0) >> COL_SHIFT;
353 col[8*7] = (a0) >> COL_SHIFT;
354 col[8*1] = (a1) >> COL_SHIFT;
355 col[8*6] = (a1) >> COL_SHIFT;
356 col[8*2] = (a2) >> COL_SHIFT;
357 col[8*5] = (a2) >> COL_SHIFT;
358 col[8*3] = (a3) >> COL_SHIFT;
359 col[8*4] = (a3) >> COL_SHIFT;
362 col[8*0] = (a0 + b0) >> COL_SHIFT;
363 col[8*7] = (a0 - b0) >> COL_SHIFT;
364 col[8*1] = (a1 + b1) >> COL_SHIFT;
365 col[8*6] = (a1 - b1) >> COL_SHIFT;
366 col[8*2] = (a2 + b2) >> COL_SHIFT;
367 col[8*5] = (a2 - b2) >> COL_SHIFT;
368 col[8*3] = (a3 + b3) >> COL_SHIFT;
369 col[8*4] = (a3 - b3) >> COL_SHIFT;
375 static void inline idctSparse2Col (int16_t * col)
377 int a0, a1, a2, a3, b0, b1, b2, b3;
378 col[0] += (1<<(COL_SHIFT-1))/W4;
438 col[8*0] = (a0 + b0) >> COL_SHIFT;
439 col[8*7] = (a0 - b0) >> COL_SHIFT;
440 col[8*1] = (a1 + b1) >> COL_SHIFT;
441 col[8*6] = (a1 - b1) >> COL_SHIFT;
442 col[8*2] = (a2 + b2) >> COL_SHIFT;
443 col[8*5] = (a2 - b2) >> COL_SHIFT;
444 col[8*3] = (a3 + b3) >> COL_SHIFT;
445 col[8*4] = (a3 - b3) >> COL_SHIFT;
449 /* If all rows but the first one are zero after row transformation,
450 all rows will be identical after column transformation. */
451 static inline void idctCol2(int16_t *col)
455 uint64_t *lcol = (uint64_t *) col;
457 for (i = 0; i < 8; ++i) {
458 int a0 = col[0] + (1 << (COL_SHIFT - 1)) / W4;
461 col[0] = a0 >> COL_SHIFT;
467 lcol[ 2] = l; lcol[ 3] = r;
468 lcol[ 4] = l; lcol[ 5] = r;
469 lcol[ 6] = l; lcol[ 7] = r;
470 lcol[ 8] = l; lcol[ 9] = r;
471 lcol[10] = l; lcol[11] = r;
472 lcol[12] = l; lcol[13] = r;
473 lcol[14] = l; lcol[15] = r;
477 void simple_idct (short *block)
487 idctRowCondDC(block);
492 nNonZero+= idctRowCondZ(block + i*8);
506 block[i+56]= (W4*block[i] + (1<<(COL_SHIFT-1))) >> COL_SHIFT;
511 int index= nonZero[0]*8;
514 int bias= W4*block[i] + (1<<(COL_SHIFT-1));
515 int c= block[i + index];
516 block[i ]= (c*coeff[index ] + bias) >> COL_SHIFT;
517 block[i+8 ]= (c*coeff[index+1] + bias) >> COL_SHIFT;
518 block[i+16]= (c*coeff[index+2] + bias) >> COL_SHIFT;
519 block[i+24]= (c*coeff[index+3] + bias) >> COL_SHIFT;
520 block[i+32]= (c*coeff[index+4] + bias) >> COL_SHIFT;
521 block[i+40]= (c*coeff[index+5] + bias) >> COL_SHIFT;
522 block[i+48]= (c*coeff[index+6] + bias) >> COL_SHIFT;
523 block[i+56]= (c*coeff[index+7] + bias) >> COL_SHIFT;
526 /* else if(nNonZero==2)
528 int index1= nonZero[0]*8;
529 int index2= nonZero[1]*8;
532 int bias= W4*block[i] + (1<<(COL_SHIFT-1));
533 int c1= block[i + index1];
534 int c2= block[i + index2];
535 block[i ]= (c1*coeff[index1 ] + c2*coeff[index2 ] + bias) >> COL_SHIFT;
536 block[i+8 ]= (c1*coeff[index1+1] + c2*coeff[index2+1] + bias) >> COL_SHIFT;
537 block[i+16]= (c1*coeff[index1+2] + c2*coeff[index2+2] + bias) >> COL_SHIFT;
538 block[i+24]= (c1*coeff[index1+3] + c2*coeff[index2+3] + bias) >> COL_SHIFT;
539 block[i+32]= (c1*coeff[index1+4] + c2*coeff[index2+4] + bias) >> COL_SHIFT;
540 block[i+40]= (c1*coeff[index1+5] + c2*coeff[index2+5] + bias) >> COL_SHIFT;
541 block[i+48]= (c1*coeff[index1+6] + c2*coeff[index2+6] + bias) >> COL_SHIFT;
542 block[i+56]= (c1*coeff[index1+7] + c2*coeff[index2+7] + bias) >> COL_SHIFT;
548 idctSparse2Col(block + i);
550 #elif defined(ARCH_ALPHA)
551 int rowsZero = 1; /* all rows except row 0 zero */
552 int rowsConstant = 1; /* all rows consist of a constant value */
554 for (i = 0; i < 8; i++) {
555 int sparseness = idctRowCondDC(block + 8 * i);
557 if (i > 0 && sparseness > 0)
565 } else if (rowsConstant) {
566 uint64_t *lblock = (uint64_t *) block;
568 idctSparseCol(block);
569 for (i = 0; i < 8; i++) {
570 uint64_t v = (uint16_t) block[i * 8];
579 for (i = 0; i < 8; i++)
580 idctSparseCol(block + i);
584 idctRowCondDC(block + i*8);
587 idctSparseCol(block + i);